A hydrogen compressor condition monitoring method based on cluster offset correction

Abstract

The present invention belongs to the field of hydrogen compressor monitoring technology, and in particular relates to a hydrogen compressor state monitoring method based on cluster offset correction. The state monitoring method includes: for any thermodynamic parameter of the hydrogen compressor: according to the window parameters of the current sliding window, obtaining the thermodynamic data of the current sliding window; calculating the statistical data of the current sliding window according to the thermodynamic data of the current sliding window; the window parameters, thermodynamic data and statistical data of the current sliding window together constitute the current sliding window information; according to the sliding window information of each thermodynamic parameter between adjacent monitoring moments, each clustering algorithm outputs the cluster offset correction result corresponding to the current monitoring moment; and calculates the weight coefficient of each clustering algorithm at the current monitoring moment; the clustering algorithm weight coefficient corresponding to the same cluster offset correction result is accumulated into the score of the corresponding cluster offset correction result, and the cluster offset correction result with the highest score is used as the hydrogen compressor operating state at the current monitoring moment. The present invention can improve the accuracy of hydrogen compressor operating state monitoring.

Description

Technical Field

[0001] The present invention belongs to the technical field of hydrogen compressor monitoring, and in particular relates to a hydrogen compressor state monitoring method based on cluster offset correction. Background Art

[0002] Hydrogen compressors (hereinafter referred to as "hydrogen compressors"), as the core boosting equipment for hydrogen refueling stations and hydrogen filling, are crucial for the widespread application of hydrogen energy. Hydrogen compressors are designed to operate at high pressures, reaching over 90 MPa. The compressors also contain complex, intertwined gas, water, and oil circuits, and numerous valves.

[0003] Due to the influence of assembly process, transportation bumps, on-site operating environment, etc., even if the hydrogen compressor passes the factory inspection, it may still malfunction during actual operation; furthermore, as time goes by, the performance of the hydrogen compressor itself deteriorates, and the internal parts age or are damaged, which will not only affect the operating efficiency of the hydrogen compressor, but also pose a threat to the personal safety of the technical personnel on site.

[0004] In addition to regularly inspecting and maintaining key components in the hydrogen compressor, technicians also use a clustering algorithm to map the condition of the hydrogen compressor based on the exhaust pressure collected by the sensor during operation, that is, to obtain the current operating conditions of the hydrogen compressor, so as to determine whether there is any problem with the current hydrogen compressor.

[0005] The existing technology is based on a single acquisition parameter, such as exhaust pressure, and then uses a single clustering algorithm to determine which cluster the current hydrogen compressor condition will fall into. The corresponding cluster label is then output as the current operating condition of the hydrogen compressor:

[0006] ① Various parameters of the hydrogen compressor influence each other, which means that the existing technology can only judge the working condition of the hydrogen compressor at the current moment based on this single thermodynamic parameter, but cannot accurately predict the working condition of the hydrogen compressor in the future.

[0007] ② When the clustering algorithm does not have the ability to autonomously generate cluster labels (that is, the cluster labels are completely set by technicians, and the clustering algorithm will not generate new clusters and corresponding cluster labels): If a new operating condition occurs during the actual operation of the hydrogen compressor, and any set cluster label is not compatible with the current operating condition of the hydrogen compressor, then the current clustering algorithm will still output a cluster label set by the technician. That is, this clustering algorithm has poor flexibility and cannot output a truly accurate hydrogen compressor operating condition.

[0008] ③ When the clustering algorithm has the ability to autonomously generate cluster labels (that is, the clustering algorithm will autonomously generate new clusters and corresponding cluster labels based on the cluster labels set by technicians): As the number of samples of the collected parameters increases, the cluster labels will continue to increase, or even grow explosively. Technicians can directly take corresponding measures on the hydrogen compressor for the set cluster labels; however, new cluster labels require technicians to verify and analyze before confirming whether measures need to be taken on the hydrogen compressor and what measures to take; and the technicians' verification and analysis are lagging, which greatly increases the workload of technicians. In actual applications, after verification and analysis by technicians, it was found that: some new cluster labels are completely wrong; some new cluster labels, although correct, can completely belong to a certain set cluster label, and only a small number of new cluster labels are accurate and different from the set cluster labels.

[0009] ④ At a certain monitoring moment, the clustering algorithm will calculate several clusters based on the collected parameters. The cluster labels of these clusters are several possible operating conditions of the current hydrogen compressor. The clustering algorithm then selects a cluster label from these clusters as the current hydrogen compressor operating condition. Therefore, the accuracy of the clusters calculated by the clustering algorithm will directly affect the clustering algorithm's judgment on the current hydrogen compressor operating condition. Among these clusters, some clusters are obtained by fission and offset of the clusters corresponding to the hydrogen compressor operating condition at the previous monitoring moment. Such clusters should not be calculated by the clustering algorithm at the current monitoring moment. For the convenience of description, such clusters are recorded as offset clusters. In other words, the appearance of offset clusters will directly reduce the accuracy of the clustering algorithm's judgment on the current hydrogen compressor operating condition.

[0010] Therefore, how to improve the accuracy of hydrogen compressor operating status monitoring has become a difficult problem that needs to be solved urgently in the field of hydrogen compressor monitoring. Summary of the Invention

[0011] The purpose of the present invention is to overcome the deficiencies of the above-mentioned prior art and provide a hydrogen compressor state monitoring method based on cluster offset correction, which can improve the accuracy of hydrogen compressor operating state monitoring.

[0012] To achieve the above object, the present invention adopts the following technical solutions:

[0013] A method for monitoring the state of a hydrogen compressor based on cluster offset correction includes the following steps:

[0014] S1, for any thermodynamic parameter of the hydrogen compressor: according to the window parameter of the current sliding window, obtain the thermodynamic data of the current sliding window; the sliding windows are arranged in chronological order;

[0015] S2, calculating the statistical data of the current sliding window based on the thermodynamic data of the current sliding window; the window parameters, thermodynamic data and statistical data of the current sliding window together constitute the current sliding window information;

[0016] S3, based on the sliding window information of each thermodynamic parameter between the current monitoring moment and the previous monitoring moment, each clustering algorithm outputs the cluster offset correction result corresponding to the current monitoring moment; and simultaneously calculates the weight coefficient of each clustering algorithm at the current monitoring moment;

[0017] S4, accumulating the clustering algorithm weight coefficients corresponding to the same cluster offset correction results as the scores of the corresponding cluster offset correction results, and then taking the cluster offset correction result with the highest score as the hydrogen compressor operating status at the current monitoring moment.

[0018] Preferably, S2' is further included after S2: S2' calculates the window parameters of the next sliding window according to the statistical data and window parameters of the current sliding window.

[0019] Preferably, when the current hydrogen compressor is monitored for the first time, or during the first operation after the current hydrogen compressor replaces one or more key components, a new monitoring cycle is entered for any thermodynamic parameter of the current hydrogen compressor, and the window parameters of the first sliding window in the new monitoring cycle are initial values; the window parameters include the window length and the sliding step; each sliding window contains only a number of thermodynamic data of one thermodynamic parameter arranged in time.

[0020] Preferably, S2' also includes the following:

[0021] Remember the current sliding window The window length is , t is a positive integer, the current sliding window The previous sliding window of , then the next sliding window Window length :

[0022] ; ;

[0023] ;

[0024] ; ;

[0025] in, Indicates the current sliding window The window length adjustment amount; Indicates the next sliding window The window length adjustment amount; is the first adjustment coefficient; is the second adjustment factor; Represents the weight coefficient of the i-th clustering algorithm. A total of n clustering algorithms are used, that is, 1≤i≤n; Represents a sliding window Intra-cluster entropy in the i-th clustering algorithm; Indicates that from the sliding window Start, including sliding window Count the historical maximum entropy values of m sliding windows forward, where m is a positive integer and m≤t; Represents a sliding window The mean vector of The modulus length is the sliding window With sliding window The difference between the average values of the thermodynamic data, The direction of the sliding window The average value of thermodynamic data points to the sliding window The average value of thermodynamic data; express 2-norm of ; is the maximum value function; Represents a sliding window The intra-cluster entropy in the i-th clustering algorithm, Represents a sliding window Intra-cluster entropy in the i-th clustering algorithm; Indicates that from the sliding window Start, including sliding window Inside, count m sliding windows forward, and the intra-cluster entropy in the corresponding clustering algorithm is obtained when i changes from 1 to n; represents covariance; represents variance; Indicates the time constant of the hydrogen compressor system; Indicates the current sliding window The sampling period of thermodynamic data is determined by the acquisition frequency of the corresponding sensor; Indicates the current sliding window The maximum permissible drift rate of the thermodynamic parameters; Indicates the current sliding window Rated values of thermodynamic parameters.

[0026] Preferably, the next sliding window Sliding step length for:

[0027] ; ; ;

[0028] in, Indicates the next sliding window The overlap ratio; Indicates the current sliding window The overlap ratio; Indicates the next sliding window The attenuation coefficient; Indicates the target overlap ratio; Indicates the total number of sliding windows that need to transition; and All are known quantities determined by the operating status of the hydrogen compressor at the last monitoring moment; Indicates that the next sliding window is included The remaining number of sliding windows that are included and need to transition, , Indicates that the current sliding window is included The remaining number of sliding windows that are included and need to be transitioned.

[0029] Preferably, S3 further includes the following:

[0030] N different clustering algorithms are used, some of which do not have the ability to autonomously generate cluster labels, while others do. Sliding window information of various thermodynamic parameters between the current monitoring moment and the previous monitoring moment is fed into each clustering algorithm. Each clustering algorithm calculates the cluster at the current monitoring moment and then performs offset correction, so that each clustering algorithm outputs the cluster offset correction result corresponding to the hydrogen compressor at the current monitoring moment.

[0031] At the same time, calculate the weight coefficient of each clustering algorithm at the current monitoring moment , 1≤i≤n and i is a positive integer: if the current monitoring moment is the first monitoring moment in the current monitoring cycle, the weight coefficients of each clustering algorithm at the current monitoring moment are obtained based on solving the multi-objective optimization function; if the current monitoring moment is not the first monitoring moment in the current monitoring cycle, the weight coefficients of each clustering algorithm at the previous monitoring moment are used to obtain the weight coefficients of each clustering algorithm at the current monitoring moment.

[0032] Preferably, the sliding window information of each thermodynamic parameter between the current monitoring moment and the previous monitoring moment is respectively fed into each clustering algorithm, each clustering algorithm calculates the cluster cluster at the current monitoring moment and then performs offset correction, so that each clustering algorithm outputs the cluster offset correction result corresponding to the hydrogen compressor at the current monitoring moment, further comprising the following sub-steps:

[0033] S31, the sliding window information of each thermodynamic parameter between the current monitoring time R and the previous monitoring time Q is respectively sent to each clustering algorithm, and the v clusters calculated by a clustering algorithm at the current monitoring time R are recorded: ,..., ,..., ,in, represents the u-th cluster calculated by the current clustering algorithm at the current monitoring time, 1≤u≤v and u and v are both positive integers;

[0034] S32, record the cluster offset correction result of the current clustering algorithm at the last monitoring time Q as , respectively calculate the correction results from cluster offset To cluster ,..., ,..., Time-varying distance of:

[0035] ;

[0036] Transmission Plan :

[0037] ;

[0038] in, Indicates the correction result from the cluster offset To cluster The time-varying distance of Indicates that from the sliding window To Sliding Window The time-varying distance of Indicates that the current clustering algorithm calculates the cluster offset correction result The sliding window corresponding to the thermodynamic parameter used at the last monitoring moment Q, the sliding window The left boundary of is on the left side of the previous monitoring moment Q, and the sliding window The time difference between the left boundary and the previous monitoring moment Q is the smallest in the corresponding thermodynamic parameter sliding window; Indicates that the current clustering algorithm calculates the cluster clusters The thermodynamic parameters used at the current monitoring time R correspond to the sliding window, sliding window The left boundary of is on the left side of the current monitoring time Q, and the sliding window The time difference between the left boundary of and the current monitoring moment R is the smallest in the corresponding thermodynamic parameter sliding window; Represents a sliding window The kth thermodynamic data in chronological order; Represents a sliding window The kth thermodynamic data in chronological order; represents the time series attenuation coefficient; Representing thermodynamic data The sampling time of Representing thermodynamic data Sampling time; 1≤k≤min(a,b) and k, a and b are all positive integers, a represents the sliding window The number of thermodynamic data included, b represents the sliding window the amount of thermodynamic data included; represents the transmission coefficient; The maximum infimum of the transmission coefficient is ; Represents a sliding window Normalized density of thermodynamic data; Represents a sliding window Normalized density of thermodynamic data; Represents thermodynamic data points With thermodynamic data points The spatial distance between

[0039] S33, respectively calculate the offset correction results from the cluster To cluster ,..., ,..., The transition probability is:

[0040] ;

[0041] in, Indicates the correction result from the cluster offset To cluster The transition probability of represents the temperature parameter; represents the fth cluster calculated by the current clustering algorithm at the current monitoring moment;

[0042] S34, performing offset correction on the v clusters calculated by the current clustering algorithm at the current monitoring time:

[0043] If the transition probability If the probability is less than the first threshold TP1, the cluster is determined to be is the valid cluster calculated by the current clustering algorithm at the current monitoring moment;

[0044] If the transition probability If the probability is above the first threshold TP1, the cluster is determined to be The offset cluster calculated by the current clustering algorithm at the current monitoring time; reset the offset cluster of the current clustering algorithm at the current monitoring time: replace the cluster label of the offset cluster with the cluster offset correction result Finally, the clusters with the same cluster label are merged and the corresponding transition probabilities are accumulated;

[0045] S35, the cluster label with the largest transfer probability after offset correction is used as the cluster offset correction result of the current clustering algorithm at the current monitoring time R Output.

[0046] Preferably, obtaining the weight coefficients of each clustering algorithm at the current monitoring moment based on solving a multi-objective optimization function also includes the following: constructing a multi-objective optimization function F and then solving it to obtain the value of the weight coefficient of each clustering algorithm at the current monitoring moment:

[0047] ;

[0048] ;

[0049] The constraints are: ;

[0050] in, is the second constant parameter; Represents the weight coefficient corresponding to the i-th clustering algorithm; represents the sum of squared errors of the i-th clustering algorithm; represents the first weight matrix, which is a column vector consisting of weight coefficients; represents the second weight matrix, which is the first weight matrix The transpose of is a row vector; Represents the first weight matrix and correlation matrix The quadratic form of is the regularization parameter.

[0051] Preferably, the weight coefficients of the clustering algorithms at the current monitoring moment are obtained based on the weight coefficients of the clustering algorithms at the previous monitoring moment, and the following contents are also included: the current monitoring moment is recorded as , the last monitoring time is , z≥2 and z is a positive integer, then the current monitoring time The weight coefficient of the i-th clustering algorithm for:

[0052] ;

[0053] in, Indicates the last monitoring moment The weight coefficient of the i-th clustering algorithm; Indicates the last monitoring moment The j-th clustering algorithm weight coefficient, 1≤j≤n and j is a positive integer; represents the learning rate; Represents the global distribution at the last monitoring moment; represents the distribution of the i-th clustering algorithm at the last monitoring moment; represents the distribution of the j-th clustering algorithm at the last monitoring moment; represents JS divergence; Representation distribution With global distribution JS divergence; Representation distribution With global distribution JS divergence.

[0054] Preferably, S5 is also included after S4: S5, if the cluster offset correction result obtained by the clustering algorithm with the ability to autonomously generate cluster labels does not become the cluster offset correction result with the highest score, then the cluster offset correction result obtained by the clustering algorithm with the ability to autonomously generate cluster labels will be recorded as a set to be verified. After the technicians regularly verify and analyze the set to be verified, if the cluster offset correction result obtained by a clustering algorithm in the set to be verified is considered correct by the technicians for Ψ consecutive times and is different from the cluster label set by the technicians, then the technicians will increase the weight coefficient of the corresponding clustering algorithm.

[0055] The beneficial effects of the present invention are:

[0056] (1) The present invention can improve the accuracy of monitoring the operating status of the hydrogen compressor.

[0057] (2) A single parameter cannot effectively monitor and warn the equipment. The hydrogen compressor status monitoring method of the present invention does not directly process a single type of data collected by the sensor, but uses the sliding window as the unit of data collection to cluster the sliding window information between two adjacent monitoring moments. For any thermodynamic parameter, the thermodynamic data in a sliding window are time-series, and the adjacent sliding windows are also time-series (new sliding windows can only be created with the passage of time); there is also some overlap of thermodynamic data between adjacent sliding windows. This allows the clustering algorithm to effectively mine and learn the potential relationships and correlations between adjacent sliding windows when processing sliding window information with time-series, thereby improving the accuracy of cluster labels obtained by each clustering algorithm in the present invention.

[0058] (3) In the hydrogen compressor status monitoring method of the present invention, a sliding window only contains one type of thermodynamic parameter. The sliding windows of different thermodynamic parameters in the same time period overlap due to the overlap in the time when the sensor collects data, which causes the sliding windows to overlap in time. The clustering algorithm in the present invention is not based on isolated thermodynamic data at a certain moment; rather, it is based on sliding window information that has time series, temporal correlation, and overlapping thermodynamic data. In addition, there is temporal overlap between the sliding windows of different thermodynamic parameters. The clustering algorithm will also mine and learn the potential relationship and correlation between these sliding window information of different thermodynamic parameters that overlap in time. This enables the clustering algorithms in the present invention to make a more accurate prediction of the current working conditions of the hydrogen compressor in the future, thereby improving the accuracy of the cluster offset correction results output by each clustering algorithm; this also makes the several clusters calculated in the clustering algorithm more accurate.

[0059] (4) The window length and sliding step size of the sliding window are directly related to the sample data input to each clustering algorithm and indirectly related to the accuracy of the cluster offset correction results output by each clustering algorithm. In each monitoring cycle of the present invention, the window length and sliding step size are automatically adjusted based on the previous sliding window, the previous sliding windows and the cluster entropy obtained by each clustering algorithm based on the previous sliding windows. No manual intervention is required. The technician only needs to set the window parameters (i.e., the initial values) of the first sliding window in each monitoring cycle, which greatly reduces labor costs.

[0060] (5) The present invention not only has a high degree of automation in adjusting the window length and sliding step size of the sliding window during the entire monitoring process, but also balances the computational overhead and storage resource usage in the process of processing the sliding window information. It also enables the clustering algorithm to better mine and learn the potential relationships and correlations between the sliding windows, allowing each clustering algorithm to better process the sliding window information and effectively perceive the long-term slowly deteriorating monitoring objects.

[0061] (6) In the present invention, the transition probability calculated based on the time-varying distance is used to determine whether there is an offset cluster in the several clusters calculated in a clustering algorithm at the current monitoring moment. If there is, the offset cluster is reset to complete the offset correction of the several clusters calculated in a clustering algorithm, thereby avoiding the occurrence of the offset cluster interfering with the clustering algorithm's judgment on the current operating status of the hydrogen compressor, and improving the accuracy of the judgment of the current operating status of the hydrogen compressor in each clustering algorithm.

[0062] (7) The present invention's determination of valid clusters and offset clusters and reset of offset clusters not only conforms to the state change trend during the working process of the hydrogen compressor, but also reflects the inheritance of the cluster label at the previous monitoring moment, thereby improving the generalization ability of the clustering algorithm.

[0063] (8) The hydrogen compressor status monitoring method of the present invention automatically updates the monitoring cycle, which is reflected in the following: when the current hydrogen compressor is monitored for the first time, or during the first operation after the current hydrogen compressor replaces one or more key components, a new monitoring cycle is entered for any thermodynamic parameter of the current hydrogen compressor. Even if it is a hydrogen compressor of the same model, it has passed the test before leaving the factory and has certain pre-factory test data; however, the process of a hydrogen compressor being transported, installed on site and then officially put into operation is the period monitored by the hydrogen compressor status monitoring method of the present invention, that is, the first time a hydrogen compressor is monitored. Therefore, before the first monitoring of a hydrogen compressor, some parts of the hydrogen compressor may still be damaged or the installation may not be compatible due to transportation and on-site installation. The same is true for the first operation after the current hydrogen compressor replaces one or more key components. Both situations mean that the hydrogen compressor as a whole needs to go through the running-in period, stabilization period and degradation period again. In addition, in each monitoring cycle, the window length and sliding step are based on the previous sliding window, so the window length and sliding step of the new sliding window in these two cases cannot be obtained based on the previous sliding window, so we need to let the hydrogen compressor in these two cases enter a new monitoring cycle.

[0064] (9) The update of the window length and sliding step size of the sliding window is related to the changes in the frequency of sample input and the amount of sample data for the subsequent clustering algorithm. The hydrogen compressor status monitoring method of the present invention automatically updates the monitoring cycle and automatically adjusts the window length and sliding step size of the sliding window within each monitoring cycle to make the frequency of sample input and the amount of sample data adaptive to the subsequent clustering algorithm processing process, further improving the efficiency of the entire monitoring process and the accuracy of the cluster offset correction results output by the clustering algorithm; therefore, the hydrogen compressor status monitoring method of the present invention can better adapt to each hydrogen compressor it monitors.

[0065] (10) The hydrogen compressor status monitoring method of the present invention does not use a single clustering algorithm to monitor the operating status of the hydrogen compressor, but uses multiple clustering algorithms for joint monitoring, and these clustering algorithms include not only clustering algorithms with the ability to autonomously generate cluster labels, but also clustering algorithms without the ability to autonomously generate cluster labels; the cluster offset correction results obtained by each of these clustering algorithms may be partially the same or completely different, but the present invention will ultimately use the cluster offset correction result with the highest score as the operating status of the hydrogen compressor at the current monitoring moment. The cluster offset correction result with the highest score is also the cluster offset correction result of most clustering algorithms. After random inspection by technical personnel, they are basically the operating status of the hydrogen compressor at the current monitoring moment.

[0066] (11) The score of the cluster offset correction result is obtained by accumulating the weight coefficients of the clustering algorithms corresponding to the same cluster offset correction result. The weight coefficients in the hydrogen compressor state monitoring method of the present invention also change with the monitoring cycle and the monitoring time within each monitoring cycle: the weight coefficients of each clustering algorithm at the first monitoring time in each monitoring cycle are determined by solving the multi-objective optimization function; and the weight coefficients of each clustering algorithm at the remaining monitoring times in each monitoring cycle need to be obtained based on the weight coefficients of each clustering algorithm at the previous monitoring time. This ensures that the weight coefficients of each clustering algorithm will not change suddenly as the monitoring time changes; and according to technical personnel, only when several cluster offset correction results obtained by a certain clustering algorithm are all the hydrogen compressor operating status at the corresponding monitoring time, the weight coefficient of the clustering algorithm will gradually increase.

[0067] (12) The hydrogen compressor status monitoring method of the present invention also takes into account a special case with an extremely low probability: if the clustering algorithm with the ability to autonomously generate cluster labels obtains a cluster offset correction result corresponding to the current monitoring moment as a new cluster label, and the weight coefficient of the clustering algorithm used to generate this new cluster label is small, then this new cluster label cannot be used as the operating status of the hydrogen compressor at the current monitoring moment; and the content of S5 is adopted accordingly. In this way, with a small amount of manual intervention, it can be ensured that when the hydrogen compressor does encounter a situation other than the cluster label preset by the technician, the clustering algorithm with a high weight coefficient and the ability to autonomously generate cluster labels can accurately determine the operating status of the hydrogen compressor at the current monitoring moment.

[0068] (13) In the hydrogen compressor status monitoring method of the present invention, the operating status of the hydrogen compressor at the current monitoring moment can be accurately determined. The operating status of the hydrogen compressor includes not only the operating condition of the hydrogen compressor at the current monitoring moment, but also the prediction of the operating condition of the current hydrogen compressor in the future. Therefore, the number of cluster labels pre-set by the technician can be more: for example, the judgment of the hydrogen compressor operating condition at the current monitoring moment in multiple cluster labels is the same, but the prediction of the operating condition of the current hydrogen compressor in the future is different. This is directly different from the existing technology that determines the operating condition of the hydrogen compressor at the current monitoring moment based on the exhaust pressure collected by the sensor. Moreover, the monitoring method of the present invention will not be prone to misjudgment due to the technician pre-setting too many cluster results as in the existing technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 This is a flow chart of a hydrogen compressor state monitoring method based on cluster offset correction according to the present invention;

[0070] Figure 2a The outlet pressure clustering result is obtained based on the outlet pressure using the DBSCAN clustering algorithm;

[0071] Figure 2b The secondary exhaust temperature clustering result is obtained based on the secondary exhaust temperature using the DBSCAN clustering algorithm;

[0072] Figure 2c The cooler temperature clustering result obtained based on the cooler temperature using the DBSCAN clustering algorithm;

[0073] Figure 3a The outlet pressure clustering result obtained based on the outlet pressure using the OPTICS clustering algorithm;

[0074] Figure 3b The secondary exhaust temperature clustering result is obtained based on the secondary exhaust temperature using the OPTICS clustering algorithm;

[0075] Figure 3c The cooler temperature clustering results obtained based on the cooler temperature using the OPTICS clustering algorithm;

[0076] Figure 4a The outlet pressure clustering result obtained based on the outlet pressure using the GMM clustering algorithm;

[0077] Figure 4b The secondary exhaust temperature clustering result is obtained based on the secondary exhaust temperature using the GMM clustering algorithm;

[0078] Figure 4c The cooler temperature clustering result obtained based on the cooler temperature using the GMM clustering algorithm;

[0079] Figure 5a The outlet pressure clustering result obtained based on the outlet pressure using the hydrogen compressor state monitoring method of the present invention;

[0080] Figure 5b The secondary exhaust temperature clustering result obtained based on the secondary exhaust temperature using the hydrogen compressor state monitoring method of the present invention;

[0081] Figure 5c This is the cooler temperature clustering result obtained based on the cooler temperature using the hydrogen compressor state monitoring method of the present invention. DETAILED DESCRIPTION

[0082] In order to make the technical solution of the present invention clearer and more specific, the present invention is clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Ordinary technicians in this field, without making any creative work, make equivalent substitutions for the technical features of the technical solution of the present invention and solutions derived from conventional reasoning all fall within the scope of protection of the present invention.

[0083] like Figure 1FIG. 1 is a flow chart of a method for monitoring a hydrogen compressor state based on cluster offset correction according to the present invention, comprising the following steps:

[0084] S1, for any thermodynamic parameter of the hydrogen compressor: according to the window parameter of the current sliding window, obtain the thermodynamic data of the current sliding window; the sliding windows are arranged in chronological order;

[0085] S2, calculating the statistical data of the current sliding window based on the thermodynamic data of the current sliding window; the window parameters, thermodynamic data and statistical data of the current sliding window together constitute the current sliding window information;

[0086] S3, based on the sliding window information of each thermodynamic parameter between the current monitoring moment and the previous monitoring moment, each clustering algorithm outputs the cluster offset correction result corresponding to the current monitoring moment; and simultaneously calculates the weight coefficient of each clustering algorithm at the current monitoring moment;

[0087] S4, accumulating the clustering algorithm weight coefficients corresponding to the same cluster offset correction results as the scores of the corresponding cluster offset correction results, and then taking the cluster offset correction result with the highest score as the hydrogen compressor operating status at the current monitoring moment.

[0088] After S2 also include S2´:

[0089] S2', calculates the window parameters of the next sliding window based on the statistical data and window parameters of the current sliding window.

[0090] In S1, the following are also included:

[0091] The thermodynamic parameters of the hydrogen compressor include: intake pressure, exhaust pressure, interstage pressure, compressor flow, intake temperature, first-stage exhaust temperature before cooling, first-stage exhaust temperature after cooling, second-stage exhaust temperature before cooling, second-stage exhaust temperature after cooling, water inlet temperature, return water temperature, lubricating oil pressure, lubricating oil temperature, first-stage cylinder pressure, second-stage cylinder pressure, first-stage bearing temperature, second-stage bearing temperature, and cooler temperature.

[0092] The key components of the hydrogen compressor include the diaphragm assembly, valve assembly, crankshaft connecting rod mechanism, sealing system, cooling system, drive motor and control system. Among them, the diaphragm assembly is the core power transmission component, and any changes in its diaphragm material, thickness, number of layers, etc. are considered to be replacement of the diaphragm assembly; the valve assembly includes the intake valve / exhaust valve, and any changes in the valve plate material, spring stiffness, valve gap, etc. are considered to be replacement of the valve assembly; the crankshaft connecting rod mechanism is the core of power conversion, and any changes in its clearance, counterweight, etc. are considered to be replacement of the crankshaft connecting rod mechanism; the sealing system includes static seals and dynamic seals, and any changes in its packing material and size are considered to be replacement of the sealing system; changes in the properties of the cooling medium in the cooling system are considered to be replacement of the cooling system; changes in the motor efficiency, power, PID control parameters, etc. in the drive motor and control system are considered to be replacement of the drive motor and control system.

[0093] When monitoring a hydrogen compressor for the first time, or during the first operation after replacing one or more key components, a new monitoring cycle begins for any thermodynamic parameter of the compressor. The window parameters of the first sliding window in the new monitoring cycle are all initial values. Window parameters include window length and sliding step size. The initial values of window parameters are set by technicians and can vary for different thermodynamic parameters.

[0094] In addition to the above situation, no matter how many times the current hydrogen compressor is restarted after being shut down, for any thermodynamic parameter of the hydrogen compressor, it is in the same monitoring cycle. In the same monitoring cycle, for a certain thermodynamic parameter, assuming that the last sliding window of the current hydrogen compressor during its last operation is the 51st sliding window in the current cycle, then the first sliding window after the current hydrogen compressor is restarted is the 52nd sliding window in the current cycle.

[0095] Each sliding window contains a number of thermodynamic data of only one thermodynamic parameter arranged in time. There are some overlaps of thermodynamic data between adjacent sliding windows of the same thermodynamic parameter.

[0096] The sliding direction of the sliding window is unique and synchronized with the time series. Each sliding step is a time interval. When the sliding window completes a sliding step, it becomes a new sliding window. Each time the sliding window slides, its sliding step is the sliding step specified in the window parameters of the current sliding window.

[0097] For the same thermodynamic parameter, the window length and window sliding step size directly affect the amount of thermodynamic data overlap between adjacent sliding windows. The window length directly determines the amount of thermodynamic data in the sliding window.

[0098] Thermodynamic data are collected regularly by corresponding sensors, and the collection frequency of the sensors is generally fixed.

[0099] S2 also includes the following:

[0100] The current sliding window statistical data includes: the mean value of thermodynamic data, the variance of thermodynamic data, the standard deviation of thermodynamic data, the median of thermodynamic data, the first quartile of thermodynamic data, the third quartile of thermodynamic data, the interquartile range of thermodynamic data, the skewness of thermodynamic data, the kurtosis of thermodynamic data, the maximum value of thermodynamic data, the minimum value of thermodynamic data, and the coefficient of variation of thermodynamic data.

[0101] The following are also included in S2´:

[0102] Remember the current sliding window The window length is , t is a positive integer, the current sliding window The previous sliding window of , then the next sliding window Window length :

[0103] ; ;

[0104] ;

[0105] ; ;

[0106] in, Indicates the current sliding window The window length adjustment amount; Indicates the next sliding window The window length adjustment amount; is the first adjustment coefficient; is the second adjustment factor; represents the weight coefficient of the i-th clustering algorithm. In the present invention, a total of n clustering algorithms are used, 1≤i≤n; Represents a sliding window Intra-cluster entropy in the i-th clustering algorithm; Indicates that from the sliding window Start (including sliding window ) Count the historical maximum entropy value of m sliding windows forward, where m is a positive integer and m≤t; Represents a sliding window The mean vector of The modulus length is the sliding window With sliding window The difference between the average values of the thermodynamic data, The direction of the sliding window The average value of thermodynamic data points to the sliding window The average value of thermodynamic data; express 2-norm of ; is the maximum value function; Represents a sliding window The intra-cluster entropy in the i-th clustering algorithm, Represents a sliding window Intra-cluster entropy in the i-th clustering algorithm; Indicates that from the sliding window Start (including sliding window ) Count m sliding windows forward and get the intra-cluster entropy of the corresponding clustering algorithm when i changes from 1 to n; represents covariance; represents variance; Indicates the time constant of the hydrogen compressor system; Indicates the current sliding window The sampling period of thermodynamic data is determined by the acquisition frequency of the corresponding sensor; Indicates the current sliding window The maximum allowable drift rate of the thermodynamic parameters is set by technicians based on experience; Indicates the current sliding window Rated values of thermodynamic parameters.

[0107] If 1=t<m, that is, from the sliding window Start (including sliding window ) There are only t sliding windows in the past, then ; If 2=t<m, then ; If 3 = t < m, then .

[0108] Optionally, calculate the next sliding window Sliding step length : , where γ is the first constant parameter, which is defined by technicians. In this embodiment, γ=0.2.

[0109] Optionally, calculate the next sliding window Sliding step length :

[0110] ; ; ;

[0111] in, Indicates the next sliding window The overlap ratio; Indicates the current sliding window The overlap ratio; Indicates the next sliding window The attenuation coefficient; Indicates the target overlap ratio; Indicates the total number of sliding windows that need to transition; and They are all determined by the operating state of the hydrogen compressor at the last monitoring moment and are all known quantities. In the present invention, technicians will divide the target overlap ratios and their corresponding total number of sliding windows to be transitioned into several levels according to the operating state of the hydrogen compressor in advance; Indicates that the next sliding window is included The remaining number of sliding windows that are included and need to transition, , Indicates that the current sliding window is included The remaining number of sliding windows that are included and need to be transitioned.

[0112] The present invention determines the next sliding window Sliding step length In the process of ": Based on the smooth transition algorithm of state transfer, the state transfer equation is constructed, that is, and ; Timely adjustment of the sliding step size of the next sliding window is to timely adjust the amount of thermodynamic data overlapping between adjacent sliding windows; and by continuously adjusting the transition of several sliding step sizes, that is, gradually changing the sliding step sizes of several sliding windows, a large stage-by-stage adjustment of the sliding step size can be achieved, avoiding the excessive change in the sliding step sizes of adjacent sliding windows, thereby affecting the subsequent clustering algorithm's mining and learning of the potential relationships and correlations between adjacent sliding windows.

[0113] Sliding Window This is the first sliding window.

[0114] In this example, if the sliding window The average value of the thermodynamic data is greater than the sliding window The average value of the thermodynamic data is The direction is positive, otherwise it is negative.

[0115] The window length is adjusted every time the sliding window slides. After completing one sliding, the window length is also adjusted, and the next sliding window is formed. The window length is The window length of the sliding window can be adjusted by adjusting the right edge of the sliding window; the left edge of the sliding window does not need to be adjusted, and the distance between the left edges of adjacent sliding windows is a sliding step.

[0116] In the calculation of intra-cluster entropy, if the clustering algorithm is Kmeans++ clustering algorithm, hierarchical clustering algorithm, DBSCAN clustering algorithm, GMM clustering algorithm, OPTICS clustering algorithm, etc., the intra-cluster entropy based on distance is calculated; for spectral clustering algorithm, the intra-cluster entropy based on similarity is calculated.

[0117] In S3, the following are also included:

[0118] N different clustering algorithms are used, some of which do not have the ability to autonomously generate cluster labels, while others do. Sliding window information of various thermodynamic parameters between the current monitoring moment and the previous monitoring moment is fed into each clustering algorithm. Each clustering algorithm calculates the cluster at the current monitoring moment and then performs offset correction, so that each clustering algorithm outputs a cluster offset correction result corresponding to the hydrogen compressor at the current monitoring moment. At the current monitoring moment, one clustering algorithm calculates more than one cluster; and one clustering algorithm outputs one cluster offset correction result.

[0119] At the same time, calculate the weight coefficient of each clustering algorithm at the current monitoring moment , 1≤i≤n and i is a positive integer: if the current monitoring moment is the first monitoring moment in the current monitoring cycle, the weight coefficients of each clustering algorithm at the current monitoring moment are obtained based on solving the multi-objective optimization function; if the current monitoring moment is not the first monitoring moment in the current monitoring cycle, the weight coefficients of each clustering algorithm at the previous monitoring moment are used to obtain the weight coefficients of each clustering algorithm at the current monitoring moment.

[0120] In this embodiment, "sliding window information of each thermodynamic parameter between the current monitoring moment and the previous monitoring moment is respectively sent to each clustering algorithm". This sliding window refers to a sliding window whose right boundary is between the current monitoring moment and the previous monitoring moment.

[0121] In this embodiment, 10 different clustering algorithms are used, namely AP clustering algorithm (affinity propagation clustering algorithm), Agglomerative clustering algorithm (agglomerative hierarchical clustering algorithm), BIRCH clustering algorithm (balanced iterative reduction clustering algorithm), DBSCAN clustering algorithm (density-based spatial clustering application algorithm), GMM clustering algorithm (Gaussian mixture model clustering algorithm), Divisive clustering algorithm (divisive hierarchical clustering algorithm), Kmeans++ clustering algorithm (K-means++ clustering algorithm), MeanShift clustering algorithm (mean shift clustering algorithm), OPTICS clustering algorithm (ordered point clustering structure recognition algorithm), Spectral clustering algorithm (spectral clustering algorithm); the corresponding weight coefficients are ~ .

[0122] Among them, the Agglomerative clustering algorithm, GMM clustering algorithm, Divisive clustering algorithm, Kmeans++ clustering algorithm and Spectral clustering algorithm do not have the ability to independently generate cluster labels; the AP clustering algorithm, BIRCH clustering algorithm, DBSCAN clustering algorithm, MeanShift clustering algorithm and OPTICS clustering algorithm have the ability to independently generate cluster labels.

[0123] The method of "sending sliding window information of each thermodynamic parameter between the current monitoring moment and the previous monitoring moment to each clustering algorithm, wherein each clustering algorithm calculates the cluster at the current monitoring moment and then performs offset correction, so that each clustering algorithm outputs the cluster offset correction result corresponding to the hydrogen compressor at the current monitoring moment" further includes the following sub-steps:

[0124] S31, the sliding window information of each thermodynamic parameter between the current monitoring time R and the previous monitoring time Q is respectively sent to each clustering algorithm, and the v clusters calculated by a clustering algorithm at the current monitoring time R are recorded: ,..., ,..., ,in, It represents the u-th cluster calculated by the current clustering algorithm at the current monitoring time, 1≤u≤v, and u and v are both positive integers.

[0125] S32, record the cluster offset correction result of the current clustering algorithm at the last monitoring time Q as , respectively calculate the correction results from cluster offset To cluster ,..., ,..., Time-varying distance of:

[0126] ;

[0127] Transmission Plan :

[0128] ;

[0129] in, Indicates the correction result from the cluster offset To cluster The time-varying distance of Indicates that from the sliding window To Sliding Window The time-varying distance of Indicates that the current clustering algorithm calculates the cluster offset correction result The sliding window corresponding to the thermodynamic parameter used at the last monitoring moment Q, the sliding window The left boundary of is on the left side of the previous monitoring moment Q, and the sliding window The time difference between the left boundary and the previous monitoring moment Q is the smallest in the corresponding thermodynamic parameter sliding window; Indicates that the current clustering algorithm calculates the cluster clusters The thermodynamic parameters used at the current monitoring time R correspond to the sliding window, sliding window The left boundary of is on the left side of the current monitoring time Q, and the sliding window The time difference between the left boundary of and the current monitoring moment R is the smallest in the corresponding thermodynamic parameter sliding window; Represents a sliding window The kth thermodynamic data in chronological order; Represents a sliding window The kth thermodynamic data in chronological order; Indicates the time series attenuation coefficient, which is used to control the impact of time differences and is set by technical personnel; Representing thermodynamic data The sampling time of Representing thermodynamic data Sampling time; 1≤k≤min(a,b) and k, a and b are all positive integers, a represents the sliding window The number of thermodynamic data included, b represents the sliding window Amount of thermodynamic data included; transfer schedule is a double marginal constraint; represents the transmission coefficient; The maximum infimum of the transmission coefficient is ; Represents a sliding window Normalized density of thermodynamic data; Represents a sliding window Normalized density of thermodynamic data; Represents thermodynamic data points With thermodynamic data points The spatial distance between them is the Euclidean distance in this embodiment.

[0130] To facilitate understanding, the following explains the "v clusters calculated within a clustering algorithm at the current monitoring moment": Even if a clustering algorithm mines and learns the potential relationships and correlations between sliding windows of the same thermodynamic parameter, or between sliding windows of different thermodynamic parameters, different clusters will be obtained due to the use of different thermodynamic parameters for calculation. The v clusters calculated within a clustering algorithm represent the v possible operating states of the hydrogen compressor at the current monitoring moment, but the clustering algorithm ultimately selects only the cluster label of one cluster as the output of the "operating state of the hydrogen compressor at the current monitoring moment" determined by the current clustering algorithm.

[0131] S33, respectively calculate the offset correction results from the cluster To cluster ,..., ,..., The transition probability is:

[0132] ;

[0133] in, Indicates the correction result from the cluster offset To cluster The transition probability of Represents the temperature parameter, set by the technician, and used to control the strictness of cluster label inheritance from the previous detection moment at the current monitoring moment. The smaller it is, the more the transition probability is concentrated on the distance cluster offset correction result. The smallest cluster, The larger it is, the more evenly the distribution of transition probabilities in each cluster is; Represents the fth cluster calculated by the current clustering algorithm at the current monitoring moment.

[0134] S34, performing offset correction on the v clusters calculated by the current clustering algorithm at the current monitoring time:

[0135] If the transition probability If the probability is less than the first threshold TP1, the cluster is determined to be is the valid cluster calculated by the current clustering algorithm at the current monitoring moment;

[0136] If the transition probability If the probability is above the first threshold TP1, the cluster is determined to be The offset cluster calculated by the current clustering algorithm at the current monitoring time; reset the offset cluster of the current clustering algorithm at the current monitoring time: replace the cluster label of the offset cluster with the cluster offset correction result Finally, the clusters with the same cluster labels are merged and the corresponding transition probabilities are accumulated.

[0137] S35, the cluster label with the largest transfer probability after offset correction is used as the cluster offset correction result of the current clustering algorithm at the current monitoring time R Output.

[0138] In this embodiment, the first probability threshold TP1 is determined by technicians based on the temperature parameter To set; in this embodiment, TP1 is 20%.

[0139] For ease of understanding, let's take S34 to S35 as an example: the cluster offset correction result of clustering algorithm 1 at the last monitoring moment is Assume that the clustering algorithm 1 at the current monitoring moment calculates four clusters: 、 、 and , the corresponding cluster labels are sign1, sign2, sign3 and sign4, and their transition probabilities are calculated to be 16%, 18%, 19% and 47% respectively. Since the first probability threshold TP1 = 20%, the clusters are 、 and All of them are offset clusters. After offset correction, the only cluster labels in clustering algorithm 1 at the current monitoring time are and two clusters of sign4, cluster labels The transfer probability of the cluster label sign4 is 53%, and the transfer probability of the cluster label sign4 is 47%. As the cluster offset correction result of clustering algorithm 1 at the current monitoring time R Output.

[0140] In the present invention, the transition probability calculated based on the time-varying distance is used to determine whether there is an offset cluster in several clusters calculated in a clustering algorithm at the current monitoring moment. If so, the offset cluster is reset to complete the offset correction of several clusters calculated in a clustering algorithm, thereby avoiding the occurrence of the offset cluster interfering with the clustering algorithm's judgment on the current operating status of the hydrogen compressor, and improving the accuracy of the judgment of the current operating status of the hydrogen compressor in each clustering algorithm.

[0141] The presence of offset clusters interferes with the clustering algorithm's judgment of the current hydrogen compressor operating status in the following two aspects:

[0142] ① For clustering algorithms that do not have the ability to independently generate cluster labels, although the cluster labels of the offset clusters are also pre-set by technicians, the cluster labels of the offset clusters may be completely inconsistent with the current operating status of the hydrogen compressor (i.e., incorrect cluster labels). This not only increases the selection range of the cluster offset correction results output by the clustering algorithm at the current monitoring moment, but also causes incorrect options to appear in the selection range.

[0143] ② For clustering algorithms that have the ability to autonomously generate cluster labels, the number of offset clusters may be large, and some of the cluster labels of these offset clusters may be completely wrong, while others may belong to a certain set cluster label (for the sake of ease of description, such cluster labels are recorded as repeated cluster labels); this not only makes the clustering algorithm output the current monitoring moment cluster offset correction result with a larger selection range, but also a large number of wrong options and repeated options (i.e., repeated cluster labels) appear in the selection range.

[0144] In summary, it can be seen that for any clustering algorithm, the emergence of offset clusters reduces the efficiency and accuracy of the cluster labels output by the clustering algorithm.

[0145] During the operation of the hydrogen compressor, except for the monitoring moments before and after the damage of the components, the operating state of the hydrogen compressor will change suddenly. Generally, the operating state of the hydrogen compressor at adjacent monitoring moments with a short time interval will not change suddenly, but will only change slightly or remain unchanged. In the process of determining the valid cluster and the offset cluster, when the transition probability is less than the first probability threshold, it is considered that the current cluster is the result of the cluster offset correction at the previous monitoring moment. The corresponding clusters are split and shifted, and these shifted clusters are reset to cluster labels When the transfer probability is above the first probability threshold, it is considered that the current cluster is indeed consistent with the cluster offset correction result. The corresponding clusters are significantly different and represent a possible operating state of the hydrogen compressor at the current monitoring moment, so they should be retained as valid clusters. Therefore, the present invention's determination of valid and offset clusters, as well as the resetting of offset clusters, not only conforms to the state change trend of the hydrogen compressor during operation, but also reflects the inheritance of the cluster label at the previous monitoring moment, improving the generalization ability of the clustering algorithm.

[0146] In the present invention, any clustering algorithm takes the cluster label with the largest transfer probability after offset correction as the cluster offset correction result of the current clustering algorithm at the current monitoring time R. The reason for output is that the greater the transfer probability, the more likely the cluster label is the current operating state of the hydrogen compressor. Subsequent verification by technical personnel has shown that S31 to S35 have indeed improved the accuracy of the cluster labels output by each clustering algorithm.

[0147] "Obtaining the weight coefficients of each clustering algorithm at the current monitoring moment based on solving the multi-objective optimization function" also includes the following:

[0148] Based on Lyapunov stability theory, a multi-objective optimization function F is constructed and solved to obtain the weight coefficients of each clustering algorithm at the current monitoring moment:

[0149] ;

[0150] ;

[0151] The constraints are: ;

[0152] in, is the second constant parameter. In this embodiment =0.001; Represents the weight coefficient corresponding to the i-th clustering algorithm; represents the sum of squared errors of the i-th clustering algorithm; Represents the first weight matrix, which is also a column vector consisting of weight coefficients; represents the second weight matrix, which is the first weight matrix The transpose of is also a row vector; Represents the first weight matrix and correlation matrix The quadratic form of is the regularization parameter.

[0153] Correlation Matrix Used to describe the relationship between different clustering algorithms; regularization parameter It is set by technical personnel to control the impact of correlation between clustering algorithms.

[0154] In this embodiment, the multi-objective optimization function F is converted into a convex optimization using the CVXPY tool, and then the built-in solvers ECOS and SCS are called to solve it.

[0155] "Acquiring the weight coefficients of each clustering algorithm at the current monitoring moment based on the weight coefficients of each clustering algorithm at the previous monitoring moment" also includes the following:

[0156] The current monitoring time is recorded as , the last monitoring time is , z≥2 and z is a positive integer, then the current monitoring time The weight coefficient of the i-th clustering algorithm for:

[0157] ;

[0158] in, Indicates the last monitoring moment The weight coefficient of the i-th clustering algorithm; Indicates the last monitoring moment The j-th clustering algorithm weight coefficient; 1≤j≤n and j is a positive integer; Represents the learning rate, which is set by technicians to control the speed of updating the algorithm weight coefficients. When it increases, it means that the algorithm adjusts the weight coefficient more actively to adapt to the distribution faster; Represents the global distribution at the last monitoring moment. The global distribution at the last monitoring moment can be obtained based on the thermodynamic data recorded in the current monitoring period and the weight coefficients of each algorithm. It is a known quantity. represents the distribution of the i-th clustering algorithm at the last monitoring moment; represents the distribution of the j-th clustering algorithm at the last monitoring moment; represents JS divergence; Representation distribution With global distribution JS divergence, used to measure and The difference in distribution With global distribution The more similar, the The larger the value; Representation distribution With global distribution JS divergence, used to measure and The difference in distribution With global distribution The more similar, the The larger the value.

[0159] After S4, S5 is also included:

[0160] S5. If the cluster offset correction result obtained by the clustering algorithm with the ability to autonomously generate cluster labels does not become the cluster offset correction result with the highest score, then the cluster offset correction result obtained by the clustering algorithm with the ability to autonomously generate cluster labels will be recorded as a verification set, and the technicians will regularly verify and analyze the verification set. If the cluster offset correction result obtained by a clustering algorithm in the verification set is considered correct by the technicians for Ψ consecutive times and is different from the cluster label set by the technicians, the technicians will increase the weight coefficient of the corresponding clustering algorithm.

[0161] In this embodiment, Ψ=5.

[0162] In the existing technology, a single type of data collected by the sensor is directly processed. Because the data at the current time point is clustered to obtain a cluster label, the corresponding cluster label is output as the current operating condition of the hydrogen compressor. Therefore, the existing technology can generally only be used to determine the current operating condition of a specific hydrogen compressor. However, this determination of the current operating condition of a hydrogen compressor is also affected by the selected clustering algorithm itself. This influence is directly reflected in the reduced accuracy of the judgment of the current operating condition of the hydrogen compressor. We have already explained this in the background technology and will not repeat it here.

[0163] Suppose a technician sets cluster label 1 as "Hydrogen compressor continues to operate normally" and cluster label 2 as "Hydrogen compressor currently operates normally, but in a low-pressure state." Cluster label 1 not only includes the judgment of the current operating condition, "Hydrogen compressor operates normally," but also includes a prediction of the current operating condition of the hydrogen compressor for a period of time in the future, "Continued normal operation." Cluster label 2 not only includes the judgment of the current operating condition, "Hydrogen compressor continues to operate normally," but also includes a prediction of the current operating condition of the hydrogen compressor for a period of time in the future, "Low-pressure operation." However, because existing clustering algorithms (regardless of whether they have the ability to autonomously generate cluster labels) perform clustering processing on data at the current point in time, they cannot accurately predict the current operating condition of the hydrogen compressor for a period of time in the future (for example, the current hydrogen compressor is indeed "in a low-pressure state" for a period of time in the future, but the existing technology determines that it will "continue to operate normally" for a period of time in the future). Therefore, even if the current operating state of the hydrogen compressor should more accurately fall into the cluster containing cluster label 2, the existing clustering algorithm will still place it in cluster label 1.

[0164] A single parameter cannot effectively monitor and provide early warnings for equipment. The hydrogen compressor status monitoring method of the present invention does not directly process a single type of data collected by the sensor. Instead, it uses sliding windows as the unit of data collection and clusters the sliding window information between two adjacent monitoring moments. For any thermodynamic parameter, the thermodynamic data within a sliding window is time-series, and so are the adjacent sliding windows (new sliding windows are only created with the passage of time). There is also some overlap in thermodynamic data between adjacent sliding windows. This allows the clustering algorithm, when processing time-series sliding window information, to effectively mine and learn the potential relationships and correlations between adjacent sliding windows, thereby improving the accuracy of the cluster labels obtained by the various clustering algorithms in the present invention.

[0165] In the hydrogen compressor status monitoring method of the present invention, a sliding window only contains one type of thermodynamic parameter. The sliding windows of different thermodynamic parameters in the same time period overlap due to the overlap in the time when the sensor collects data, which causes the sliding windows to overlap in time. The clustering algorithm in the present invention is not based on isolated thermodynamic data at a certain moment; rather, it is based on sliding window information that has time sequence, temporal correlation, and overlapping thermodynamic data. In addition, there is temporal overlap between the sliding windows of different thermodynamic parameters. The clustering algorithm will also mine and learn the potential relationships and correlations between these sliding window information of different thermodynamic parameters that overlap in time. This allows the various clustering algorithms in the present invention to make more accurate predictions about the current operating conditions of the hydrogen compressor in the future, thereby improving the accuracy of the cluster offset correction results output by each clustering algorithm; this also makes the several cluster clusters calculated in the clustering algorithm more accurate.

[0166] From the above analysis, we can see that the window length and sliding step size of the sliding window directly affect the sample data input to each clustering algorithm and indirectly affect the accuracy of the cluster offset correction results output by each clustering algorithm. In each monitoring cycle, the window length and sliding step size are automatically adjusted based on the previous sliding window, the previous sliding windows, and the intra-cluster entropy obtained by each clustering algorithm based on these previous sliding windows. This eliminates the need for manual intervention; technicians only need to set the window parameters (i.e., initial values) for the first sliding window in each monitoring cycle, significantly reducing labor costs. A window length that is too long or a sliding step size that is too short will significantly increase the computational overhead of the clustering algorithm when processing sliding window information and occupy a large amount of storage resources. While a window length that is too short or a sliding step size that is too long will reduce computational overhead and storage resource usage, the amount of thermodynamic data overlapping between adjacent sliding windows will also be greatly reduced or even eliminated. The distribution of sliding windows with temporal overlap between different thermodynamic parameter sliding windows will also change, significantly hindering the clustering algorithm's ability to mine and learn the potential relationships and correlations between sliding windows and even affecting the accuracy of the cluster offset correction results output by each clustering algorithm. Therefore, the present invention's adjustment of the sliding window length and sliding step size throughout the entire monitoring process is not only highly automated, but also balances computational overhead and storage resource usage when processing sliding window information. It also enables the clustering algorithm to better mine and learn the potential relationships and correlations between sliding windows, allowing each clustering algorithm to better process sliding window information and effectively perceive long-term, slowly degrading monitoring objects.

[0167] The hydrogen compressor status monitoring method of the present invention will automatically update the monitoring cycle, which is reflected in: when the current hydrogen compressor is monitored for the first time, or during the first operation after the current hydrogen compressor replaces one or more key components, a new monitoring cycle is entered for any thermodynamic parameter of the current hydrogen compressor. Even if it is a hydrogen compressor of the same model, it has passed the test before leaving the factory and has certain pre-factory test data; however, the process of a hydrogen compressor being transported, installed on site and then officially put into operation is the period monitored by the hydrogen compressor status monitoring method of the present invention, that is, the first time a hydrogen compressor is monitored. Therefore, before the first monitoring of a hydrogen compressor, the hydrogen compressor may still be damaged due to transportation and on-site installation, or the installation may not be compatible; the same is true for the first operation of the current hydrogen compressor after replacing one or more key components; both situations mean that the hydrogen compressor as a whole needs to go through the running-in period, stabilization period and deterioration period again. In addition, in each monitoring cycle, the window length and sliding step are based on the previous sliding window, so the window length and sliding step of the new sliding window in these two cases cannot be obtained based on the previous sliding window, so we need to let the hydrogen compressor in these two cases enter a new monitoring cycle.

[0168] Based on the above analysis, it can be seen that the update of the window length and sliding step size of the sliding window is related to the frequency of sample input and the change of sample data volume for the subsequent clustering algorithm. The hydrogen compressor status monitoring method of the present invention automatically updates the monitoring cycle and automatically adjusts the window length and sliding step size of the sliding window in each monitoring cycle to make the frequency of sample input and the sample data volume adaptive to the subsequent clustering algorithm processing process, further improving the efficiency of the entire monitoring process and the accuracy of the cluster offset correction results output by the clustering algorithm; therefore, the hydrogen compressor status monitoring method of the present invention can better adapt to each hydrogen compressor it monitors. For ease of understanding, for example: after entering a new monitoring cycle, the hydrogen compressor status monitoring method of the present invention needs to frequently input samples and increase the amount of input sample data in order to better understand the current situation of the hydrogen compressor in the current monitoring cycle; it has been verified by technical personnel that the present invention does indeed make the window length of the sliding window grow slowly and the sliding step of the sliding window is relatively short within a period of time after the monitoring cycle is updated (i.e., the running-in period), so as to meet the need for frequent sample input and increase the amount of input sample data during the running-in period of the current monitoring cycle; and in the middle period of the new monitoring cycle (i.e., the stable period), the window length and the sliding step of the sliding window are indeed increased significantly, while there is no need to frequently input samples or input a large amount of sample data during the stable period of the current monitoring cycle (reflected in the reduction of overlapping thermodynamic data between adjacent sliding windows and the increase in the sliding step of the sliding window); subsequently, as the clustering algorithm determines the operating status of the hydrogen compressor, the frequency of sample input and the amount of input sample data will be increased because the hydrogen compressor enters the degradation period of the current monitoring cycle, which will not be repeated here.

[0169] The hydrogen compressor status monitoring method of the present invention does not use a single clustering algorithm to monitor the operating status of the hydrogen compressor, but uses multiple clustering algorithms for joint monitoring, and these clustering algorithms include not only clustering algorithms with the ability to autonomously generate cluster labels, but also clustering algorithms that do not have the ability to autonomously generate cluster labels; the cluster offset correction results obtained by each of these clustering algorithms may be partially the same or may be completely different, but the present invention will ultimately use the cluster offset correction result with the highest score as the operating status of the hydrogen compressor at the current monitoring moment. The cluster offset correction result with the highest score is also the cluster offset correction result of most clustering algorithms. After random inspection by technical personnel, they are basically the operating status of the hydrogen compressor at the current monitoring moment.

[0170] The score of the cluster offset correction result is obtained by accumulating the clustering algorithm weight coefficients corresponding to the same cluster offset correction result, and the weight coefficients in the hydrogen compressor status monitoring method of the present invention also change with the monitoring cycle and the monitoring time within each monitoring cycle: the weight coefficients of each clustering algorithm at the first monitoring time in each monitoring cycle are determined by solving the multi-objective optimization function; and the weight coefficients of each clustering algorithm at the remaining monitoring times in each monitoring cycle need to be obtained based on the weight coefficients of each clustering algorithm at the previous monitoring time. This ensures that the weight coefficients of each clustering algorithm will not mutate as they change with the monitoring time; and it has been verified by technical personnel that only when several cluster offset correction results obtained continuously by a certain clustering algorithm are all the hydrogen compressor operating status at the corresponding monitoring time, the weight coefficient of the clustering algorithm will gradually increase.

[0171] Considering the extremely rare special case: if a clustering algorithm with the ability to autonomously generate cluster labels obtains a new cluster label as the cluster offset correction result corresponding to the current monitoring moment, and the weight coefficient of the clustering algorithm that generates this new cluster label is small, then this new cluster label cannot be used as the operating status of the hydrogen compressor at the current monitoring moment. Therefore, the present invention records the cluster labels obtained by all clustering algorithms with the ability to autonomously generate cluster labels as a verification set. That is, in S5, after the technician regularly verifies and analyzes the verification set, if the cluster offset correction result obtained by a clustering algorithm in the verification set is considered correct by the technician for Ψ consecutive times, and the cluster offset correction result is different from the cluster label set by the technician, the technician increases the weight coefficient of the corresponding clustering algorithm. This means that although the cluster offset correction result output by the clustering algorithm is not the cluster label pre-set by the technician, it has high accuracy. In this way, with a small amount of manual intervention, it can be ensured that when the hydrogen compressor actually encounters a situation other than the cluster label preset by the technician, the clustering algorithm with a high weight coefficient and the ability to autonomously generate cluster labels can accurately determine the operating status of the hydrogen compressor at the current monitoring moment.

[0172] In the hydrogen compressor status monitoring method of the present invention, the operating status of the hydrogen compressor at the current monitoring moment can be accurately determined. The operating status of the hydrogen compressor includes not only the operating conditions of the hydrogen compressor at the current monitoring moment, but also the prediction of the operating conditions of the current hydrogen compressor in the future. Therefore, the number of cluster labels pre-set by technicians can be more: for example, the judgment of the hydrogen compressor operating conditions at the current monitoring moment is the same in multiple cluster labels, but the prediction of the operating conditions of the current hydrogen compressor in the future is different. This is directly different from the prior art, which judges the operating conditions of the hydrogen compressor at the current monitoring moment based on the exhaust pressure collected by the sensor. Moreover, the monitoring method of the present invention will not be prone to misjudgment due to the technicians pre-setting too many cluster results as in the prior art.

[0173] like Figure 2a to Figure 2c 、 Figure 3a to Figure 3c 、 Figure 4a to Figure 4c As shown, the horizontal axis is the serial number of the data segment, which is the corresponding clustering result obtained by technicians using existing technology based on different single thermodynamic parameters. Figure 2a to Figure 2c The corresponding clustering results obtained using the DBSCAN clustering algorithm according to different thermodynamic parameters are: Figure 2a This is the outlet pressure clustering result based on the outlet pressure: the data during the stable period is marked in green, the purple indicates abnormal operating conditions, and a small number of normal operating conditions are misidentified; Figure 2b The following is the clustering result of the secondary exhaust temperature based on the secondary exhaust temperature: the data during the stable period is marked in purple, the green is the step interval, and the yellow is the abnormal operating condition. It can be seen that some mutation points cannot be identified and are mixed in the same type of operating conditions; Figure 2c This is the cooler temperature clustering result based on the cooler temperature. The data during the stable period is marked in purple, and the yellow identification is abnormal operating conditions. The abnormal conditions in the purple interval are not effectively identified, and the yellow is also an incorrect identification, with a very high error rate. Figure 3a to Figure 3c The corresponding clustering results obtained using the OPTICS clustering algorithm according to different thermodynamic parameters are: Figure 3a This is the outlet pressure clustering result based on the outlet pressure: cyan, green, and most purple are correctly identified as normal operating conditions, yellow is correctly identified as deviated operating conditions, but severe abnormal conditions are identified as purple conditions, which is an incorrect identification. Figure 3b The secondary exhaust temperature clustering result is obtained based on the secondary exhaust temperature: normal operating conditions are identified as green and yellow, and operating conditions in the abnormal range are marked as green and purple. A large number of misidentifications also occur; Figure 3c The cooler temperature clustering result obtained based on the cooler temperature: the algorithm generally marks abnormal operating conditions as green and blue, and the other colors are normal operating conditions. However, the green mark representing the abnormality still appears in some normal operating conditions, and the algorithm accuracy is still insufficient. Figure 4a to Figure 4c The corresponding clustering results obtained using the GMM clustering algorithm according to different thermodynamic parameters are: Figure 4a The output pressure clustering result based on the output pressure is shown in Figure 2: The normal operating conditions are generally marked in yellow, purple, and green, with a small amount of cyan. The abnormal range is mostly marked in green, with a small amount of green. However, green and cyan are distributed in both normal and abnormal ranges, and the accuracy needs to be improved. Figure 4b The secondary exhaust temperature clustering result is obtained based on the secondary exhaust temperature: the secondary exhaust temperature is only roughly identified as blue, yellow, and a small amount of purple. It can be seen that the boundary between abnormal and normal operating conditions cannot be distinguished; Figure 4cThe cooler temperature clustering results based on the cooler temperature effectively identify the abnormal interval as the purple interval, but the green and blue markers are marked in both operating conditions, resulting in misidentification. This shows that the existing technology can identify the deviation of the hydrogen compressor outlet pressure at the current monitoring time, but cannot distinguish the degree; it is insufficient to identify abnormal operating conditions in the steep rising edge of the secondary exhaust temperature; the recognition accuracy of the cooler temperature is not high, and the variable temperature data is not effectively recognized; and the different thermodynamic parameters are independent of each other.

[0174] like Figure 5a to Figure 5c As shown, the horizontal axis is the serial number of the data segment, which is the corresponding clustering result (i.e., cluster offset correction result) obtained by integrating multiple thermodynamic parameters in the hydrogen compressor state monitoring method of the present invention. Figure 5a This is the outlet pressure clustering result based on the outlet pressure: abnormal operating conditions are marked with purple, green, and blue according to different deviation degrees, and normal operating conditions are marked with yellow, cyan, and pink; Figure 5b Based on the secondary exhaust temperature clustering results and the cooler temperature clustering results, the abnormal intervals are effectively identified as green and blue. The outlet pressure clustering results show that the present invention effectively identifies the secondary exhaust temperature and cooler temperature anomalies to automatically distinguish different pressure deviation levels. It can be seen that in the scenario where the clustering clusters are offset, the present invention can effectively identify the rising edge of the operating condition corresponding to the deviation cluster. Figure 5c The cooler temperature clustering result is obtained based on the cooler temperature. The cooler temperature clustering result is obtained by correlating thermodynamic parameters such as the secondary exhaust temperature, the cooler temperature and the outlet pressure of the deviation level, thereby effectively identifying the variable temperature data.

[0175] For the six cluster labels pre-set by technicians, the hydrogen compressor status monitoring method of the present invention is used. Compared with the GMM clustering algorithm used in the prior art that does not have the ability to independently generate cluster labels, the same hydrogen compressor is continuously monitored for 133 hours, with an interval of 5 minutes between adjacent monitoring times. After manual verification, the accuracy of the hydrogen compressor status monitoring method of the present invention is as high as 96.18%, while the prior art is only 66.73%; and during this period, the hydrogen compressor status monitoring method of the present invention also generates a cluster offset correction result for a new cluster label as the operating status of the hydrogen compressor at the corresponding monitoring time. After verification by technicians, the operating status is correct and different from the six cluster labels pre-set by the technicians.

[0176] The technologies, shapes, and structures not described in detail in the present invention are all well-known technologies. It should also be pointed out that the above are only preferred embodiments of the present invention and are not intended to limit the present invention. The various components or steps in the embodiments of the present invention can be decomposed and / or recombined, and such decompositions and / or recombinations should be regarded as equivalent solutions of the present application and should fall within the scope of protection of the present invention.

Claims

1. A hydrogen compressor state monitoring method based on cluster offset correction, characterized in that: The following steps are involved: S1, for any thermodynamic parameter of the hydrogen compressor: according to the window parameter of the current sliding window, obtain the thermodynamic data of the current sliding window; the sliding windows are arranged in chronological order; S2, calculating the statistical data of the current sliding window based on the thermodynamic data of the current sliding window; the window parameters, thermodynamic data and statistical data of the current sliding window together constitute the current sliding window information; S3, based on the sliding window information of each thermodynamic parameter between the current monitoring moment and the previous monitoring moment, each clustering algorithm outputs the cluster offset correction result corresponding to the current monitoring moment; and simultaneously calculates the weight coefficient of each clustering algorithm at the current monitoring moment; S4, accumulating the clustering algorithm weight coefficients corresponding to the same cluster offset correction results as the scores of the corresponding cluster offset correction results, and then using the cluster offset correction result with the highest score as the hydrogen compressor operating status at the current monitoring moment; S2' is also included after S2: S2' calculates the window parameters of the next sliding window based on the statistical data and window parameters of the current sliding window; The following are also included in S2´: Remember the current sliding window The window length is , t is a positive integer, the current sliding window The previous sliding window of , then the next sliding window Window length : ； in, Indicates the current sliding window The window length adjustment amount; Indicates the next sliding window The window length adjustment amount; is the first adjustment coefficient; is the second adjustment factor; Represents the weight coefficient of the i-th clustering algorithm. A total of n clustering algorithms are used, that is, 1≤i≤n; Represents a sliding window Intra-cluster entropy in the i-th clustering algorithm; Indicates that from the sliding window Start, including sliding window Count the historical maximum entropy values of m sliding windows forward, where m is a positive integer and m≤t; Represents a sliding window The mean vector of The modulus length is the sliding window With sliding window The difference between the average values of the thermodynamic data, The direction of the sliding window The average value of thermodynamic data points to the sliding window The average value of thermodynamic data; express 2-norm of ; is the maximum value function; Represents a sliding window The intra-cluster entropy in the i-th clustering algorithm, Represents a sliding window Intra-cluster entropy in the i-th clustering algorithm; Indicates that from the sliding window Start, including sliding window Inside, count m sliding windows forward, and the intra-cluster entropy in the corresponding clustering algorithm is obtained when i changes from 1 to n; represents covariance; represents variance; Indicates the time constant of the hydrogen compressor system; Indicates the current sliding window The sampling period of thermodynamic data is determined by the acquisition frequency of the corresponding sensor; Indicates the current sliding window The maximum permissible drift rate of the thermodynamic parameters; Indicates the current sliding window Rated values of thermodynamic parameters.

2. The method for monitoring hydrogen compressor status based on cluster offset correction according to claim 1, characterized in that: When the current hydrogen compressor is monitored for the first time, or during the first operation after replacing one or more key components of the current hydrogen compressor, any thermodynamic parameter of the current hydrogen compressor enters a new monitoring cycle, and the window parameters of the first sliding window in the new monitoring cycle are initial values; The window parameters include the window length and the sliding step size; each sliding window contains a number of thermodynamic data of only one thermodynamic parameter arranged in time.

3. The method for monitoring hydrogen compressor status based on cluster offset correction according to claim 1, characterized in that: Next sliding window Sliding step length for: ， ， ； in, Indicates the next sliding window The overlap ratio; Indicates the current sliding window The overlap ratio; Indicates the next sliding window The attenuation coefficient; Indicates the target overlap ratio; Indicates the total number of sliding windows that need to transition; and All are known quantities determined by the operating status of the hydrogen compressor at the last monitoring moment; Indicates that the next sliding window is included The remaining number of sliding windows that are included and need to transition, , Indicates that the current sliding window is included The remaining number of sliding windows that are included and need to be transitioned.

4. The method for monitoring hydrogen compressor status based on cluster offset correction according to claim 1, characterized in that: In S3, the following are also included: N different clustering algorithms are used, some of which do not have the ability to autonomously generate cluster labels, while others do. Sliding window information of various thermodynamic parameters between the current monitoring moment and the previous monitoring moment is fed into each clustering algorithm. Each clustering algorithm calculates the cluster at the current monitoring moment and then performs offset correction, so that each clustering algorithm outputs the cluster offset correction result corresponding to the hydrogen compressor at the current monitoring moment. At the same time, calculate the weight coefficient of each clustering algorithm at the current monitoring moment , 1≤i≤n and i is a positive integer: if the current monitoring moment is the first monitoring moment in the current monitoring cycle, the weight coefficients of each clustering algorithm at the current monitoring moment are obtained based on solving the multi-objective optimization function; if the current monitoring moment is not the first monitoring moment in the current monitoring cycle, the weight coefficients of each clustering algorithm at the previous monitoring moment are used to obtain the weight coefficients of each clustering algorithm at the current monitoring moment.

5. The method for monitoring hydrogen compressor status based on cluster offset correction according to claim 4, characterized in that: The sliding window information of each thermodynamic parameter between the current monitoring moment and the previous monitoring moment is respectively fed into each clustering algorithm. Each clustering algorithm calculates the cluster at the current monitoring moment and then performs offset correction, so that each clustering algorithm outputs the cluster offset correction result corresponding to the hydrogen compressor at the current monitoring moment. The following sub-steps are also included: S31, the sliding window information of each thermodynamic parameter between the current monitoring time R and the previous monitoring time Q is respectively sent to each clustering algorithm, and the v clusters calculated by a clustering algorithm at the current monitoring time R are recorded: ,..., ,..., ,in, represents the u-th cluster calculated by the current clustering algorithm at the current monitoring time, 1≤u≤v and u and v are both positive integers; S32, record the cluster offset correction result of the current clustering algorithm at the last monitoring time Q as , respectively calculate the correction results from cluster offset To cluster ,..., ,..., Time-varying distance of: ； Transmission Plan : ； in, Indicates the correction result from the cluster offset To cluster The time-varying distance of Indicates that from the sliding window To Sliding Window The time-varying distance of Indicates that the current clustering algorithm calculates the cluster offset correction result The sliding window corresponding to the thermodynamic parameter used at the last monitoring moment Q, the sliding window The left boundary of is on the left side of the previous monitoring moment Q, and the sliding window The time difference between the left boundary and the previous monitoring moment Q is the smallest in the corresponding thermodynamic parameter sliding window; Indicates that the current clustering algorithm calculates the cluster clusters The thermodynamic parameters used at the current monitoring time R correspond to the sliding window, sliding window The left boundary of is on the left side of the current monitoring time Q, and the sliding window The time difference between the left boundary of and the current monitoring moment R is the smallest in the corresponding thermodynamic parameter sliding window; Represents a sliding window The kth thermodynamic data in chronological order; Represents a sliding window The kth thermodynamic data in chronological order; represents the time series attenuation coefficient; Representing thermodynamic data The sampling time of Representing thermodynamic data Sampling time; 1≤k≤min(a,b) and k, a and b are all positive integers, a represents the sliding window The number of thermodynamic data included, b represents the sliding window the amount of thermodynamic data included; represents the transmission coefficient; The maximum infimum of the transmission coefficient is ; Represents a sliding window Normalized density of thermodynamic data; Represents a sliding window Normalized density of thermodynamic data; Represents thermodynamic data points With thermodynamic data points The spatial distance between S33, respectively calculate the offset correction results from the cluster To cluster ,..., ,..., The transition probability is: ； in, Indicates the correction result from the cluster offset To cluster The transition probability of represents the temperature parameter; represents the fth cluster calculated by the current clustering algorithm at the current monitoring moment; S34, performing offset correction on the v clusters calculated by the current clustering algorithm at the current monitoring time: If the transition probability If the probability is less than the first threshold TP1, the cluster is determined to be is the valid cluster calculated by the current clustering algorithm at the current monitoring moment; If the transition probability If the probability is above the first threshold TP1, the cluster is determined to be The offset cluster calculated by the current clustering algorithm at the current monitoring time; reset the offset cluster of the current clustering algorithm at the current monitoring time: replace the cluster label of the offset cluster with the cluster offset correction result Finally, the clusters with the same cluster label are merged and the corresponding transition probabilities are accumulated; S35, the cluster label with the largest transfer probability after offset correction is used as the cluster offset correction result of the current clustering algorithm at the current monitoring time R Output.

6. The method for monitoring hydrogen compressor status based on cluster offset correction according to claim 4, characterized in that: The weight coefficients of each clustering algorithm at the current monitoring moment are obtained by solving the multi-objective optimization function, which also includes the following: constructing the multi-objective optimization function F and then solving it to obtain the value of the weight coefficient of each clustering algorithm at the current monitoring moment: ； ； The constraints are: ; in, is the second constant parameter; Represents the weight coefficient corresponding to the i-th clustering algorithm; represents the sum of squared errors of the i-th clustering algorithm; represents the first weight matrix, which is a column vector consisting of weight coefficients; represents the second weight matrix, which is the first weight matrix The transpose of is a row vector; Represents the first weight matrix and correlation matrix The quadratic form of is the regularization parameter.

7. The method for monitoring hydrogen compressor status based on cluster offset correction according to claim 5, characterized in that: Based on the weight coefficients of each clustering algorithm at the previous monitoring time, the weight coefficients of each clustering algorithm at the current monitoring time are obtained, which also includes the following: The current monitoring time is recorded as , the last monitoring time is , z≥2 and z is a positive integer, then the current monitoring time The weight coefficient of the i-th clustering algorithm for: ； in, Indicates the last monitoring moment The weight coefficient of the i-th clustering algorithm; Indicates the last monitoring moment The j-th clustering algorithm weight coefficient, 1≤j≤n and j is a positive integer; represents the learning rate; Represents the global distribution at the last monitoring moment; represents the distribution of the i-th clustering algorithm at the last monitoring moment; represents the distribution of the j-th clustering algorithm at the last monitoring moment; ) represents JS divergence; Representation distribution With global distribution JS divergence; Representation distribution With global distribution JS divergence.

8. The method for monitoring hydrogen compressor status based on cluster offset correction according to claim 1, characterized in that: S4 also includes S5: S5, if the cluster offset correction result obtained by the clustering algorithm with the ability to autonomously generate cluster labels does not become the cluster offset correction result with the highest score, then the cluster offset correction result obtained by the clustering algorithm with the ability to autonomously generate cluster labels will be recorded as a set to be verified. After the technical staff regularly verifies and analyzes the set to be verified, if the cluster offset correction result obtained by a clustering algorithm in the set to be verified is considered correct by the technical staff for Ψ consecutive times and is different from the cluster label set by the technical staff, then the technical staff will increase the weight coefficient of the corresponding clustering algorithm.

Images